Adaptation to Unknown Situations as the Holy Grail of Learning-Based\n Self-Adaptive Systems: Research Directions

Abstract

Self-adaptive systems continuously adapt to changes in their execution\nenvironment. Capturing all possible changes to define suitable behaviour\nbeforehand is unfeasible, or even impossible in the case of unknown changes,\nhence human intervention may be required. We argue that adapting to unknown\nsituations is the ultimate challenge for self-adaptive systems. Learning-based\napproaches are used to learn the suitable behaviour to exhibit in the case of\nunknown situations, to minimize or fully remove human intervention. While such\napproaches can, to a certain extent, generalize existing adaptations to new\nsituations, there is a number of breakthroughs that need to be achieved before\nsystems can adapt to general unknown and unforeseen situations. We posit the\nresearch directions that need to be explored to achieve unanticipated\nadaptation from the perspective of learning-based self-adaptive systems. At\nminimum, systems need to define internal representations of previously unseen\nsituations on-the-fly, extrapolate the relationship to the previously\nencountered situations to evolve existing adaptations, and reason about the\nfeasibility of achieving their intrinsic goals in the new set of conditions. We\nclose discussing whether, even when we can, we should indeed build systems that\ndefine their own behaviour and adapt their goals, without involving a human\nsupervisor.\n